ctubbsii commented on a change in pull request #2185:
URL: https://github.com/apache/accumulo/pull/2185#discussion_r666551581
##########
File path: server/manager/src/main/java/org/apache/accumulo/manager/Manager.java
##########
@@ -1120,6 +1122,9 @@ boolean canSuspendTablets() {
if (null != upgradeMetadataFuture) {
upgradeMetadataFuture.get();
}
+ if (null != upgradeFilesFuture) {
+ upgradeFilesFuture.get();
+ }
Review comment:
Well, usually we don't upgrade files at all, because the number of files
tends to be excessive and upgrading them at service startup doesn't scale well.
So, we support upgrading files over time in a distributed way, by reading
multiple older versions, and writing only newer versions. See RFile code. I
think we even still have some remaining code for reading MapFiles (which we
used before RFile existed). In general, we don't want to upgrade files at all
at this phase, certainly not isolated to a single machine, when we could have
thousands and thousands of files across HDFS.
In this case, I was concerned that these files being dropped depend on
metadata changes... since there could be references to these sorted WAL files
in the metadata. That would be one example. In general, though, we just don't
upgrade files at this point at all.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
